Unifying Logical and Statistical AI (2006) [pdf]

Lise Getoor have a nice talk at NIPS in 2017 on a what I think is a formalism, Probabilistic Soft Logic, that her group developed: https://youtu.be/t4k5LKCpboc . Definitely an interesting direction of research.

Lise Getoor has been doing the Lord's* work for a while now, presenting statistical relational learning work at the major machine learning conferences. She had the keynote speech at NIPS last year, supporting the use of structural bias in statistical learning:

https://www.youtube.com/watch?v=t4k5LKCpboc

Although she had to riff through a couple of very unfortunate slides quickly (identifying political affiliation of social network users is a classic task in relational learning, but the keynote came hot in the heels of the Cambridge Analytica scandal and her slides went down like a led baloon) it was a great keynote and one I was happy to see.

There's more work on combining relational, logic-based machine learning and statistical machine learning by Luc de Raedt, Kristian Kerstig and Stephen Muggleton, among many others in the Inductive Logic Programming community. There's some good pointers on this website for Lise Getoor's introductory book to the field:

https://www.cs.umd.edu/srl-book/

___________

* Lord Sauron's.

Here's a short, textual introduction to probabilistic soft logic, the formalism presented in the video:

> A Short Introduction to Probabilistic Soft Logic

http://stephenbach.net/files/kimmig-probprog12.pdf

And a longer, more formal treatment of the same:

> Hinge-Loss Markov Random Fields and Probabilistic Soft Logic

http://www.jmlr.org/papers/volume18/15-631/15-631.pdf